plawolf
Lieutenant General
Getting bored in Friday afternoon with tummy full of Chinese food, I would like to try some conspiracy thinking:
The timing of the latest DeepSeek releases.
DeepSeek-V3 was released on December 26th, 2024, which turned out to be the prelude of this whole DeepSeek drama. People were in western holiday season and had plenty of free time. Then on January 20th, 2025 DeepSeek-R1 was released, which was accidentally the inauguration day and 9 days before Chinese New Year. Also accidentally OpenAI and Meta et al. had already announced their new big releases "in early 2025", which up until then was the benchmark for closed LLM model and open LLM model. Now everyone in the world and their grannies are waiting for benchmarking new shiny releases from hyperscalers against this new kid on the block, DeepSeek. In between, there were also meaningful releases from Alibaba Qwen2.5-max, ByteDance Doubao 1.5 pro and Moonshoot Kimi.ai k1.5.
(1) Nobody is talking about Tiktok
(2) Everyone and his aunt is AI expert
(3) Small guys all over the world are coming out defending DeepSeek as an angel of open source
(4) Hyperscalers are uncharacteristically quiet in comparison to their 500B heydays prior to DeepSeek-R1 release
(5) US TV talking heads are actually mild on DeepSeek in comparison to their stance on Tiktok
(6) Some in China said it all happened due to Liang Wen Feng's conviction to open source
(7) Other in China said there some big hands behind this we-are-releasing-this-open-source-model playbook
What am I thinking now? Not much except that hats off to young engineers in China, who are more idealistic and less materialistic than being portrayed in social media. Anyhow, it has become a historical fact that, a private self-funded small tech firm in the beautiful heaven of Hangzhou has altered trajectory of generative AI development during the initial phase of the 4th industrial revolution.
My conspiracy thinking says that all this is not accident......
It would be a mistake to think DeepSeek are just being naive and idealistic here to gift their model to all of humanity.
You need to remember who they are and what their day jobs are. If anyone could have predicted the magnitude of impact DeepSeek would have on the NASAQ, it’s these guys.
They didn’t release DeepSeek fully out of some idealistic naïveté, but because it was a carefully weighed decision to both maximise their own profits from this breakthrough, as well as cause maximum damage to the strong enemy.
If they kept it as a closed book proprietary model, the market impact would have been far far smaller, barely a blip as everyone just shrugged and moved on since the market would expect the western AI oligarchs to do their thing and got DeepSeek banned in America, Europe and everywhere else they can and massively limit the market DeepSeek can operate in, which in turn will also massively limit its potential revenue. Add in the inevitable massed and sustained cyberattacks and they would be luck to make any sort of meaningful profits from it.
If they shorted NVIDIA and other obvious targets’ stocks before the release, DeepSeek could have made them billions overnight.
Given the choice between pressing send and making billions overnight or spending then next decade fighting the collective west to carve out some tiny market share to maybe make a few million, what would you choose? Kneecapping America’s collective hyper scale AI industry on the way is just a nice bonus.